Your AI workforce

Agents

A library of specialist AI agents, each with its own face. Browse them by category, chat with one directly, or open its editor to shape its profile, face, instructions, tools, models, memory and limits.

An agent is a specialist with its own position, face, instructions, model and tools. Nova, your main assistant, can hand work to agents. You can also talk to one directly, put several in a team, or use them as steps in a workflow.

Agents live in the Agency window, on the Agents tab. Open it from Agency in the sidebar, the office-building icon.

The Agents tab: every agent with its own face and country flag, grouped in categories, each card with Chat, info, edit, duplicate and delete.

The Agency window

  • The tab strip: Office, Agents, Teams, Workflows, Scheduler and Goals. Office shows your whole workforce: a live floor plan, and its org chart of departments, teams and positions.
  • The main button on the right changes with the tab: New Agent, New Team, New Workflow, New Schedule or New Goal. The round arrow beside it reloads every list.
  • Themes: every agent, team and workflow wears one gradient on its cover, its avatar and its main button, unless you upload your own images. An agent's avatar is its face, painted on that gradient.

The library

  • Agent Categories: agents grouped under headings such as Core Systems, Marketing, Sales, Finance & Accounting, Development & Engineering, Design, Legal or Healthcare & Life Sciences. Each heading shows how many agents it holds.
  • The toolbar on the right: search, a category filter, a switch between cards and a table, and a button for a new category.
  • Agent cards: a cover, the agent's face with its country flag, its name and position, a short description, its category and up to two tags. Each card has five buttons: Chat (in the agent's own colours), Agent info, edit, duplicate and delete.
  • The table lists Name, Category, Position, Tools and Model, with chat, info, edit and delete on each row.
  • Agent info opens a quick summary in place: position, category, country, description and tags.

SyntrofAI ships with over a hundred ready-made agents across nineteen categories. They behave exactly like agents you create: you can chat with them, edit them, duplicate them or remove them.

Chat with an agent

Press Chat on a card, or at the top of the agent's editor. A conversation with that agent opens inside the window.

  • Type and press Enter to send. Press Shift + Enter for a new line. Attach adds files.
  • Agent switches work mode on or off. Work mode lets it use tools and take actions. RAG appears when the agent has its own memory, and lets it draw on that memory.
  • The side list keeps that agent's past sessions. New starts a fresh one, and the arrow folds the list away.
  • Back returns to the library. Main Chat moves the same conversation to the main screen, so you can keep going with the full composer.

Create an agent

  1. Press New Agent.
  2. Fill in Agent Name and Role. Optionally add a Category, a Description, Initial AGENTS.md Instructions and Writable Folders. Leave the instructions blank and SyntrofAI writes a starter file from the name and role.
  3. Press Create. The new agent starts with a random face and theme gradient; change both in its editor.
  4. To give the agent models of its own, press edit on the new card, open Models, pick a provider and a model, and press Save changes.

The agent editor

The agent editor: sections on the left, the agent's cover and face on top, and the Profile page with its identity and face.

edit opens the agent's editor. The rail on the left groups its sections under Identity, Capabilities and Guardrails. The top of the page shows the agent's cover, face, name and position, with Chat, Duplicate and Delete. Your changes stay in the editor until you press Save changes at the bottom of the rail, which wears the agent's own gradient.

Profile

Name, Position, Category, Country (its flag shows on the agent's card) and Description. The agent's Face, described next. A Theme gradient that paints the cover, the background of the face and the Chat button. Your own Cover image, with Use gradient to go back.

Face

Face: build how the agent looks, from gender and skin to hair, eyes, eyewear and clothes, with Shuffle and Default face.

Every agent has a face, painted over its theme gradient. It is the agent's picture everywhere: its card, the chat picker and composer, the session list, its team's chips and map, and its desk in Office. Nova keeps the SyntrofAI logo.

  • Gender: Female or Male.
  • Skin and Hair colour: eight shades each.
  • Hair: seven styles for a woman (Long, Bob, Pixie, Bun, Ponytail, Curly, Afro) and eight for a man (Short, Side part, Quiff, Buzz, Curls, Afro, Shoulder length, Bald). A man can also have Facial hair: Stubble, Mustache, Goatee, Beard or Full beard.
  • Eye shape (nine shapes) and Eye colour (six).
  • Eyewear: Round glasses, Square glasses, Sunglasses or Aviators.
  • Earrings (Studs, Hoops, Drops) and Headwear (Cap, Beanie, Fedora).
  • Clothes: T-shirt, Hoodie, Shirt, Blazer or Turtleneck, in one of ten colours or Any colour you pick.

Shuffle fills in a random face. Default face goes back to the face the agent was given. Neither is kept until you press Save changes.

Instructions

Instructions: the agent's AGENTS.md, and the runtime prompt files below it.

AGENTS.md is what the agent is told before every turn: what it is, how it works and what it must never do. Below it, Runtime prompt files lists the built-in instruction files the agent runs on. Switch one off for this agent, or write an override that replaces its text for this agent only. Use default brings the original back.

Tools

Tools: every tool grouped by category, with its risk, each set to Default, On or Off.

Every tool, grouped by category, with a line on what it does. Tags mark the tools that need approval and those with high or critical risk. Set each tool to Default (the agent's standard set), On or Off. Only Final Answer and Read Result, which every agent needs in order to reply, show Always on; every other tool is your choice. The counts, a search and an All, On, Off and Default filter sit at the top.

Skills

Skills: the agent's own skills, the general library, and skills borrowed from other agents.

What the agent knows how to do. Own skills belong only to this agent: create one, upload a .zip, or edit them; an own skill wins over a library skill with the same name. General library: use every library skill, or only the ones you switch on. From other agents: borrow a skill another agent owns; it stays theirs, so their edits reach this agent too. See Skills.

Models

Models: the Primary model that does the work and the Utility model for quick background chores.

A Primary model that answers and does the real work, and a Utility model for fast background chores such as summaries and choosing tools. Each has a Provider and a Model, picked from searchable lists with logos, a Temperature from Precise to Creative, and a limit on Max output tokens. Default tier chooses which of the two the agent starts on: Primary or Utility. Its Secondary and Tertiary choices have no model to set yet.

Memory

Memory: switched on, with the agent's own private store, its storage key, and the Explorer below.

Agent memory on or off, and where memories live: Own Agent Memory (private to this agent, recommended), Default Memory (the shared store Nova uses), or a Custom Namespace you name. The Explorer browses the folders and memories the agent has. Import memories copies or moves memories in from another store: preview a folder, check the memories you want, and press Import to agent. The Recall tuning values are saved but do not change recall yet.

Knowledge

Knowledge: drop files in, and follow each one as it is learned into the agent's memory.

Drop PDF, Word, Excel, PowerPoint, Markdown, text, CSV, JSON or HTML files into the agent's knowledge. Every upload is learned into the agent's own memory. Each file shows In memory, Learning or Not learned, and the list updates by itself while files are being learned.

Budget

Budget: limits on tokens and cost, and what happens when one is reached.

Hard limits on Tokens per session, Cost per turn and Cost per session (in US dollars). Each card says limited or unlimited; 0 means no limit. With Stop chosen under When a limit is reached, the agent halts the moment a limit is hit. Tokens per turn, Warn at, Warn and Fall back are saved but do not change how the agent runs yet.

Governance

Governance: the folders this agent may write to, at the bottom.

Writable folders: the folders this agent may change, one per line. . allows everything. A custom agent can always change its own home and its knowledge folder. Require approval, Retries and Time limit are saved but not used yet.

Organize the library

  • Move an agent: drag its card onto another category.
  • Add a category: use the folder button in the toolbar.
  • Rename a category: use the pencil on its heading, or double-click the name.
  • Delete a category: use the bin on its heading. Its agents move to Uncategorized, which cannot be deleted.
  • Find an agent: the search looks at name, position, description, category, tags and tool names. Use the filter to show one category.
  • Duplicate makes a copy called "name (Copy)" with the same position, instructions, tools, look and budget. The copy gets its own fresh memory and starts on the default models.
  • Delete removes the agent and takes it out of any team it belonged to. Its memory is kept.

Agents everywhere else

  • In the main chat, pick an agent from the picker in the composer to talk to it with the full set of controls. See Chat with Nova.
  • Nova delegates to agents on its own when a task benefits from a specialist. You see the hand-off in the live work log.
  • A team is a group of agents with a coordinator.
  • Workflow Agent Task steps run one agent turn as a step. See Workflows.
  • In Coding Lab, the AI developer is chosen from the engineering agents.